NamespaceString NamespaceString::makeDropPendingNamespace(const repl::OpTime& opTime) const {
mongo::StringBuilder ss;
ss << db() << "." << dropPendingNSPrefix;
ss << opTime.getSecs() << "i" << opTime.getTimestamp().getInc() << "t" << opTime.getTerm();
ss << "." << coll();
return NamespaceString(ss.stringData().substr(0, MaxNsCollectionLen));
}
/**
* Updates the session state with the last write timestamp and transaction for that session.
*
* In the case of writes with transaction/statement id, this method will be recursively entered a
* second time for the actual write to the transactions table. Since this write does not generate an
* oplog entry, the recursion will stop at this point.
*/
void onWriteOpCompleted(OperationContext* opCtx,
const NamespaceString& nss,
Session* session,
std::vector<StmtId> stmtIdsWritten,
const repl::OpTime& lastStmtIdWriteOpTime,
Date_t lastStmtIdWriteDate) {
if (lastStmtIdWriteOpTime.isNull())
return;
if (session) {
session->onWriteOpCompletedOnPrimary(opCtx,
*opCtx->getTxnNumber(),
std::move(stmtIdsWritten),
lastStmtIdWriteOpTime,
lastStmtIdWriteDate);
}
}
/**
* Updates the session state with the last write timestamp and transaction for that session.
*
* In the case of writes with transaction/statement id, this method will be recursively entered a
* second time for the actual write to the transactions table. Since this write does not generate an
* oplog entry, the recursion will stop at this point.
*/
void onWriteOpCompleted(OperationContext* opCtx,
const NamespaceString& nss,
std::vector<StmtId> stmtIdsWritten,
const repl::OpTime& lastStmtIdWriteOpTime,
Date_t lastStmtIdWriteDate,
boost::optional<DurableTxnStateEnum> txnState) {
if (lastStmtIdWriteOpTime.isNull())
return;
const auto txnParticipant = TransactionParticipant::get(opCtx);
if (!txnParticipant)
return;
txnParticipant->onWriteOpCompletedOnPrimary(opCtx,
*opCtx->getTxnNumber(),
std::move(stmtIdsWritten),
lastStmtIdWriteOpTime,
lastStmtIdWriteDate,
txnState);
}
// Adds a wOpTime and a wElectionId field to a set of gle options
static BSONObj buildGLECmdWithOpTime(const BSONObj& gleOptions,
const repl::OpTime& opTime,
const OID& electionId) {
BSONObjBuilder builder;
BSONObjIterator it(gleOptions);
for (int i = 0; it.more(); ++i) {
BSONElement el = it.next();
// Make sure first element is getLastError : 1
if (i == 0) {
StringData elName(el.fieldName());
if (!elName.equalCaseInsensitive("getLastError")) {
builder.append("getLastError", 1);
}
}
builder.append(el);
}
opTime.append(&builder, "wOpTime");
builder.appendOID("wElectionId", const_cast<OID*>(&electionId));
return builder.obj();
}
bool run(OperationContext* txn, const string& dbname,
BSONObj& cmdObj,
int,
string& errmsg,
BSONObjBuilder& result) {
//
// Correct behavior here is very finicky.
//
// 1. The first step is to append the error that occurred on the previous operation.
// This adds an "err" field to the command, which is *not* the command failing.
//
// 2. Next we parse and validate write concern options. If these options are invalid
// the command fails no matter what, even if we actually had an error earlier. The
// reason for checking here is to match legacy behavior on these kind of failures -
// we'll still get an "err" field for the write error.
//
// 3. If we had an error on the previous operation, we then return immediately.
//
// 4. Finally, we actually enforce the write concern. All errors *except* timeout are
// reported with ok : 0.0, to match legacy behavior.
//
// There is a special case when "wOpTime" and "wElectionId" are explicitly provided by
// the client (mongos) - in this case we *only* enforce the write concern if it is
// valid.
//
// We always need to either report "err" (if ok : 1) or "errmsg" (if ok : 0), even if
// err is null.
//
LastError *le = &LastError::get(txn->getClient());
le->disable();
// Always append lastOp and connectionId
Client& c = *txn->getClient();
if (repl::getGlobalReplicationCoordinator()->getReplicationMode() ==
repl::ReplicationCoordinator::modeReplSet) {
const repl::OpTime lastOp = repl::ReplClientInfo::forClient(c).getLastOp();
if (!lastOp.isNull()) {
result.append("lastOp", lastOp.getTimestamp());
// TODO(siyuan) Add "lastOpTerm"
}
}
// for sharding; also useful in general for debugging
result.appendNumber( "connectionId" , c.getConnectionId() );
Timestamp lastTimestamp;
BSONField<Timestamp> wOpTimeField("wOpTime");
FieldParser::FieldState extracted = FieldParser::extract(cmdObj, wOpTimeField,
&lastTimestamp, &errmsg);
if (!extracted) {
result.append("badGLE", cmdObj);
appendCommandStatus(result, false, errmsg);
return false;
}
repl::OpTime lastOpTime;
bool lastOpTimePresent = extracted != FieldParser::FIELD_NONE;
if (!lastOpTimePresent) {
// Use the client opTime if no wOpTime is specified
lastOpTime = repl::ReplClientInfo::forClient(c).getLastOp();
// TODO(siyuan) Fix mongos to supply wOpTimeTerm, then parse out that value here
} else {
// TODO(siyuan) Don't use the default term after fixing mongos.
lastOpTime = repl::OpTime(lastTimestamp, repl::OpTime::kDefaultTerm);
}
OID electionId;
BSONField<OID> wElectionIdField("wElectionId");
extracted = FieldParser::extract(cmdObj, wElectionIdField,
&electionId, &errmsg);
if (!extracted) {
result.append("badGLE", cmdObj);
appendCommandStatus(result, false, errmsg);
return false;
}
bool electionIdPresent = extracted != FieldParser::FIELD_NONE;
bool errorOccurred = false;
// Errors aren't reported when wOpTime is used
if ( !lastOpTimePresent ) {
if ( le->getNPrev() != 1 ) {
errorOccurred = LastError::noError.appendSelf( result, false );
}
else {
errorOccurred = le->appendSelf( result, false );
}
}
BSONObj writeConcernDoc = cmdObj;
// Use the default options if we have no gle options aside from wOpTime/wElectionId
const int nFields = cmdObj.nFields();
bool useDefaultGLEOptions = (nFields == 1) ||
(nFields == 2 && lastOpTimePresent) ||
(nFields == 3 && lastOpTimePresent && electionIdPresent);
WriteConcernOptions writeConcern;
if (useDefaultGLEOptions) {
writeConcern = repl::getGlobalReplicationCoordinator()->getGetLastErrorDefault();
}
Status status = writeConcern.parse( writeConcernDoc );
//
// Validate write concern no matter what, this matches 2.4 behavior
//
if ( status.isOK() ) {
// Ensure options are valid for this host
status = validateWriteConcern( writeConcern );
}
if ( !status.isOK() ) {
result.append( "badGLE", writeConcernDoc );
return appendCommandStatus( result, status );
}
// Don't wait for replication if there was an error reported - this matches 2.4 behavior
if ( errorOccurred ) {
dassert( !lastOpTimePresent );
return true;
}
// No error occurred, so we won't duplicate these fields with write concern errors
dassert( result.asTempObj()["err"].eoo() );
dassert( result.asTempObj()["code"].eoo() );
// If we got an electionId, make sure it matches
if (electionIdPresent) {
if (repl::getGlobalReplicationCoordinator()->getReplicationMode() !=
repl::ReplicationCoordinator::modeReplSet) {
// Ignore electionIds of 0 from mongos.
if (electionId != OID()) {
errmsg = "wElectionId passed but no replication active";
result.append("code", ErrorCodes::BadValue);
return false;
}
}
else {
if (electionId != repl::getGlobalReplicationCoordinator()->getElectionId()) {
LOG(3) << "oid passed in is " << electionId
<< ", but our id is "
<< repl::getGlobalReplicationCoordinator()->getElectionId();
errmsg = "election occurred after write";
result.append("code", ErrorCodes::WriteConcernFailed);
return false;
}
}
}
txn->setWriteConcern(writeConcern);
txn->setMessage( "waiting for write concern" );
WriteConcernResult wcResult;
status = waitForWriteConcern( txn, lastOpTime, &wcResult );
wcResult.appendTo( writeConcern, &result );
// For backward compatibility with 2.4, wtimeout returns ok : 1.0
if ( wcResult.wTimedOut ) {
dassert( !wcResult.err.empty() ); // so we always report err
dassert( !status.isOK() );
result.append( "errmsg", "timed out waiting for slaves" );
result.append( "code", status.code() );
return true;
}
return appendCommandStatus( result, status );
}